Artificial neural network model based control

Model based control is now emerging as a technology that can offer significant benefits. However, the majority of approaches assume linearity of the process and adopt an algorithm that is far removed from the ´simple´ PI(D) structure. The use of a nonlinear model within a predictive control scheme is one means of overcoming linearity limitations but the unfamiliar structure can restrict application. However, when the system dead time is small, a PI(D) controller (including feedforward) which has an auto-tuning capability can achieve comparable performance. This paper investigates a long range predictive controller and contrasts the performance with a conventional control system. Both controllers are designed using a neural network.